fs/notify/dnotify/dnotify.c

   1 /*
   2  * Directory notifications for Linux.
   3  *
   4  * Copyright (C) 2000,2001,2002 Stephen Rothwell
   5  *
   6  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
   7  * dnotify was largly rewritten to use the new fsnotify infrastructure
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms of the GNU General Public License as published by the
  11  * Free Software Foundation; either version 2, or (at your option) any
  12  * later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  */
  19 #include <linux/fs.h>
  20 #include <linux/module.h>
  21 #include <linux/sched.h>
  22 #include <linux/dnotify.h>
  23 #include <linux/init.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/slab.h>
  26 #include <linux/fdtable.h>
  27 #include <linux/fsnotify_backend.h>
  28
  29 int dir_notify_enable __read_mostly = 1;
  30
  31 static struct kmem_cache *dnotify_struct_cache __read_mostly;
  32 static struct kmem_cache *dnotify_mark_entry_cache __read_mostly;
  33 static struct fsnotify_group *dnotify_group __read_mostly;
  34 static DEFINE_MUTEX(dnotify_mark_mutex);
  35
  36 /*
  37  * dnotify will attach one of these to each inode (i_fsnotify_mark_entries) which
  38  * is being watched by dnotify.  If multiple userspace applications are watching
  39  * the same directory with dnotify their information is chained in dn
  40  */
  41 struct dnotify_mark_entry {
  42         struct fsnotify_mark_entry fsn_entry;
  43         struct dnotify_struct *dn;
  44 };
  45
  46 /*
  47  * When a process starts or stops watching an inode the set of events which
  48  * dnotify cares about for that inode may change.  This function runs the
  49  * list of everything receiving dnotify events about this directory and calculates
  50  * the set of all those events.  After it updates what dnotify is interested in
  51  * it calls the fsnotify function so it can update the set of all events relevant
  52  * to this inode.
  53  */
  54 static void dnotify_recalc_inode_mask(struct fsnotify_mark_entry *entry)
  55 {
  56         __u32 new_mask, old_mask;
  57         struct dnotify_struct *dn;
  58         struct dnotify_mark_entry *dnentry  = container_of(entry,
  59                                                            struct dnotify_mark_entry,
  60                                                            fsn_entry);
  61
  62         assert_spin_locked(&entry->lock);
  63
  64         old_mask = entry->mask;
  65         new_mask = 0;
  66         for (dn = dnentry->dn; dn != NULL; dn = dn->dn_next)
  67                 new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
  68         entry->mask = new_mask;
  69
  70         if (old_mask == new_mask)
  71                 return;
  72
  73         if (entry->inode)
  74                 fsnotify_recalc_inode_mask(entry->inode);
  75 }
  76
  77 /*
  78  * Mains fsnotify call where events are delivered to dnotify.
  79  * Find the dnotify mark on the relevant inode, run the list of dnotify structs
  80  * on that mark and determine which of them has expressed interest in receiving
  81  * events of this type.  When found send the correct process and signal and
  82  * destroy the dnotify struct if it was not registered to receive multiple
  83  * events.
  84  */
  85 static int dnotify_handle_event(struct fsnotify_group *group,
  86                                 struct fsnotify_event *event)
  87 {
  88         struct fsnotify_mark_entry *entry = NULL;
  89         struct dnotify_mark_entry *dnentry;
  90         struct inode *to_tell;
  91         struct dnotify_struct *dn;
  92         struct dnotify_struct **prev;
  93         struct fown_struct *fown;
  94         __u32 test_mask = event->mask & ~FS_EVENT_ON_CHILD;
  95
  96         to_tell = event->to_tell;
  97
  98         spin_lock(&to_tell->i_lock);
  99         entry = fsnotify_find_mark_entry(group, to_tell);
 100         spin_unlock(&to_tell->i_lock);
 101
 102         /* unlikely since we alreay passed dnotify_should_send_event() */
 103         if (unlikely(!entry))
 104                 return 0;
 105         dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 106
 107         spin_lock(&entry->lock);
 108         prev = &dnentry->dn;
 109         while ((dn = *prev) != NULL) {
 110                 if ((dn->dn_mask & test_mask) == 0) {
 111                         prev = &dn->dn_next;
 112                         continue;
 113                 }
 114                 fown = &dn->dn_filp->f_owner;
 115                 send_sigio(fown, dn->dn_fd, POLL_MSG);
 116                 if (dn->dn_mask & FS_DN_MULTISHOT)
 117                         prev = &dn->dn_next;
 118                 else {
 119                         *prev = dn->dn_next;
 120                         kmem_cache_free(dnotify_struct_cache, dn);
 121                         dnotify_recalc_inode_mask(entry);
 122                 }
 123         }
 124
 125         spin_unlock(&entry->lock);
 126         fsnotify_put_mark(entry);
 127
 128         return 0;
 129 }
 130
 131 /*
 132  * Given an inode and mask determine if dnotify would be interested in sending
 133  * userspace notification for that pair.
 134  */
 135 static bool dnotify_should_send_event(struct fsnotify_group *group,
 136                                       struct inode *inode, __u32 mask)
 137 {
 138         struct fsnotify_mark_entry *entry;
 139         bool send;
 140
 141         /* !dir_notify_enable should never get here, don't waste time checking
 142         if (!dir_notify_enable)
 143                 return 0; */
 144
 145         /* not a dir, dnotify doesn't care */
 146         if (!S_ISDIR(inode->i_mode))
 147                 return false;
 148
 149         spin_lock(&inode->i_lock);
 150         entry = fsnotify_find_mark_entry(group, inode);
 151         spin_unlock(&inode->i_lock);
 152
 153         /* no mark means no dnotify watch */
 154         if (!entry)
 155                 return false;
 156
 157         mask = (mask & ~FS_EVENT_ON_CHILD);
 158         send = (mask & entry->mask);
 159
 160         fsnotify_put_mark(entry); /* matches fsnotify_find_mark_entry */
 161
 162         return send;
 163 }
 164
 165 static void dnotify_free_mark(struct fsnotify_mark_entry *entry)
 166 {
 167         struct dnotify_mark_entry *dnentry = container_of(entry,
 168                                                           struct dnotify_mark_entry,
 169                                                           fsn_entry);
 170
 171         BUG_ON(dnentry->dn);
 172
 173         kmem_cache_free(dnotify_mark_entry_cache, dnentry);
 174 }
 175
 176 static struct fsnotify_ops dnotify_fsnotify_ops = {
 177         .handle_event = dnotify_handle_event,
 178         .should_send_event = dnotify_should_send_event,
 179         .free_group_priv = NULL,
 180         .freeing_mark = NULL,
 181         .free_event_priv = NULL,
 182 };
 183
 184 /*
 185  * Called every time a file is closed.  Looks first for a dnotify mark on the
 186  * inode.  If one is found run all of the ->dn entries attached to that
 187  * mark for one relevant to this process closing the file and remove that
 188  * dnotify_struct.  If that was the last dnotify_struct also remove the
 189  * fsnotify_mark_entry.
 190  */
 191 void dnotify_flush(struct file *filp, fl_owner_t id)
 192 {
 193         struct fsnotify_mark_entry *entry;
 194         struct dnotify_mark_entry *dnentry;
 195         struct dnotify_struct *dn;
 196         struct dnotify_struct **prev;
 197         struct inode *inode;
 198
 199         inode = filp->f_path.dentry->d_inode;
 200         if (!S_ISDIR(inode->i_mode))
 201                 return;
 202
 203         spin_lock(&inode->i_lock);
 204         entry = fsnotify_find_mark_entry(dnotify_group, inode);
 205         spin_unlock(&inode->i_lock);
 206         if (!entry)
 207                 return;
 208         dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 209
 210         mutex_lock(&dnotify_mark_mutex);
 211
 212         spin_lock(&entry->lock);
 213         prev = &dnentry->dn;
 214         while ((dn = *prev) != NULL) {
 215                 if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
 216                         *prev = dn->dn_next;
 217                         kmem_cache_free(dnotify_struct_cache, dn);
 218                         dnotify_recalc_inode_mask(entry);
 219                         break;
 220                 }
 221                 prev = &dn->dn_next;
 222         }
 223
 224         spin_unlock(&entry->lock);
 225
 226         /* nothing else could have found us thanks to the dnotify_mark_mutex */
 227         if (dnentry->dn == NULL)
 228                 fsnotify_destroy_mark_by_entry(entry);
 229
 230         fsnotify_recalc_group_mask(dnotify_group);
 231
 232         mutex_unlock(&dnotify_mark_mutex);
 233
 234         fsnotify_put_mark(entry);
 235 }
 236
 237 /* this conversion is done only at watch creation */
 238 static __u32 convert_arg(unsigned long arg)
 239 {
 240         __u32 new_mask = FS_EVENT_ON_CHILD;
 241
 242         if (arg & DN_MULTISHOT)
 243                 new_mask |= FS_DN_MULTISHOT;
 244         if (arg & DN_DELETE)
 245                 new_mask |= (FS_DELETE | FS_MOVED_FROM);
 246         if (arg & DN_MODIFY)
 247                 new_mask |= FS_MODIFY;
 248         if (arg & DN_ACCESS)
 249                 new_mask |= FS_ACCESS;
 250         if (arg & DN_ATTRIB)
 251                 new_mask |= FS_ATTRIB;
 252         if (arg & DN_RENAME)
 253                 new_mask |= FS_DN_RENAME;
 254         if (arg & DN_CREATE)
 255                 new_mask |= (FS_CREATE | FS_MOVED_TO);
 256
 257         return new_mask;
 258 }
 259
 260 /*
 261  * If multiple processes watch the same inode with dnotify there is only one
 262  * dnotify mark in inode->i_fsnotify_mark_entries but we chain a dnotify_struct
 263  * onto that mark.  This function either attaches the new dnotify_struct onto
 264  * that list, or it |= the mask onto an existing dnofiy_struct.
 265  */
 266 static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnentry,
 267                      fl_owner_t id, int fd, struct file *filp, __u32 mask)
 268 {
 269         struct dnotify_struct *odn;
 270
 271         odn = dnentry->dn;
 272         while (odn != NULL) {
 273                 /* adding more events to existing dnofiy_struct? */
 274                 if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
 275                         odn->dn_fd = fd;
 276                         odn->dn_mask |= mask;
 277                         return -EEXIST;
 278                 }
 279                 odn = odn->dn_next;
 280         }
 281
 282         dn->dn_mask = mask;
 283         dn->dn_fd = fd;
 284         dn->dn_filp = filp;
 285         dn->dn_owner = id;
 286         dn->dn_next = dnentry->dn;
 287         dnentry->dn = dn;
 288
 289         return 0;
 290 }
 291
 292 /*
 293  * When a process calls fcntl to attach a dnotify watch to a directory it ends
 294  * up here.  Allocate both a mark for fsnotify to add and a dnotify_struct to be
 295  * attached to the fsnotify_mark.
 296  */
 297 int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 298 {
 299         struct dnotify_mark_entry *new_dnentry, *dnentry;
 300         struct fsnotify_mark_entry *new_entry, *entry;
 301         struct dnotify_struct *dn;
 302         struct inode *inode;
 303         fl_owner_t id = current->files;
 304         struct file *f;
 305         int destroy = 0, error = 0;
 306         __u32 mask;
 307
 308         /* we use these to tell if we need to kfree */
 309         new_entry = NULL;
 310         dn = NULL;
 311
 312         if (!dir_notify_enable) {
 313                 error = -EINVAL;
 314                 goto out_err;
 315         }
 316
 317         /* a 0 mask means we are explicitly removing the watch */
 318         if ((arg & ~DN_MULTISHOT) == 0) {
 319                 dnotify_flush(filp, id);
 320                 error = 0;
 321                 goto out_err;
 322         }
 323
 324         /* dnotify only works on directories */
 325         inode = filp->f_path.dentry->d_inode;
 326         if (!S_ISDIR(inode->i_mode)) {
 327                 error = -ENOTDIR;
 328                 goto out_err;
 329         }
 330
 331         /* expect most fcntl to add new rather than augment old */
 332         dn = kmem_cache_alloc(dnotify_struct_cache, GFP_KERNEL);
 333         if (!dn) {
 334                 error = -ENOMEM;
 335                 goto out_err;
 336         }
 337
 338         /* new fsnotify mark, we expect most fcntl calls to add a new mark */
 339         new_dnentry = kmem_cache_alloc(dnotify_mark_entry_cache, GFP_KERNEL);
 340         if (!new_dnentry) {
 341                 error = -ENOMEM;
 342                 goto out_err;
 343         }
 344
 345         /* convert the userspace DN_* "arg" to the internal FS_* defines in fsnotify */
 346         mask = convert_arg(arg);
 347
 348         /* set up the new_entry and new_dnentry */
 349         new_entry = &new_dnentry->fsn_entry;
 350         fsnotify_init_mark(new_entry, dnotify_free_mark);
 351         new_entry->mask = mask;
 352         new_dnentry->dn = NULL;
 353
 354         /* this is needed to prevent the fcntl/close race described below */
 355         mutex_lock(&dnotify_mark_mutex);
 356
 357         /* add the new_entry or find an old one. */
 358         spin_lock(&inode->i_lock);
 359         entry = fsnotify_find_mark_entry(dnotify_group, inode);
 360         spin_unlock(&inode->i_lock);
 361         if (entry) {
 362                 dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 363                 spin_lock(&entry->lock);
 364         } else {
 365                 fsnotify_add_mark(new_entry, dnotify_group, inode);
 366                 spin_lock(&new_entry->lock);
 367                 entry = new_entry;
 368                 dnentry = new_dnentry;
 369                 /* we used new_entry, so don't free it */
 370                 new_entry = NULL;
 371         }
 372
 373         rcu_read_lock();
 374         f = fcheck(fd);
 375         rcu_read_unlock();
 376
 377         /* if (f != filp) means that we lost a race and another task/thread
 378          * actually closed the fd we are still playing with before we grabbed
 379          * the dnotify_mark_mutex and entry->lock.  Since closing the fd is the
 380          * only time we clean up the mark entries we need to get our mark off
 381          * the list. */
 382         if (f != filp) {
 383                 /* if we added ourselves, shoot ourselves, it's possible that
 384                  * the flush actually did shoot this entry.  That's fine too
 385                  * since multiple calls to destroy_mark is perfectly safe, if
 386                  * we found a dnentry already attached to the inode, just sod
 387                  * off silently as the flush at close time dealt with it.
 388                  */
 389                 if (dnentry == new_dnentry)
 390                         destroy = 1;
 391                 goto out;
 392         }
 393
 394         error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
 395         if (error) {
 396                 /* if we added, we must shoot */
 397                 if (dnentry == new_dnentry)
 398                         destroy = 1;
 399                 goto out;
 400         }
 401
 402         error = attach_dn(dn, dnentry, id, fd, filp, mask);
 403         /* !error means that we attached the dn to the dnentry, so don't free it */
 404         if (!error)
 405                 dn = NULL;
 406         /* -EEXIST means that we didn't add this new dn and used an old one.
 407          * that isn't an error (and the unused dn should be freed) */
 408         else if (error == -EEXIST)
 409                 error = 0;
 410
 411         dnotify_recalc_inode_mask(entry);
 412 out:
 413         spin_unlock(&entry->lock);
 414
 415         if (destroy)
 416                 fsnotify_destroy_mark_by_entry(entry);
 417
 418         fsnotify_recalc_group_mask(dnotify_group);
 419
 420         mutex_unlock(&dnotify_mark_mutex);
 421         fsnotify_put_mark(entry);
 422 out_err:
 423         if (new_entry)
 424                 fsnotify_put_mark(new_entry);
 425         if (dn)
 426                 kmem_cache_free(dnotify_struct_cache, dn);
 427         return error;
 428 }
 429
 430 static int __init dnotify_init(void)
 431 {
 432         dnotify_struct_cache = KMEM_CACHE(dnotify_struct, SLAB_PANIC);
 433         dnotify_mark_entry_cache = KMEM_CACHE(dnotify_mark_entry, SLAB_PANIC);
 434
 435         dnotify_group = fsnotify_obtain_group(DNOTIFY_GROUP_NUM,
 436                                               0, &dnotify_fsnotify_ops);
 437         if (IS_ERR(dnotify_group))
 438                 panic("unable to allocate fsnotify group for dnotify\n");
 439         return 0;
 440 }
 441
 442 module_init(dnotify_init)